Multi-objective prediction and optimization of working condition parameters for piston rod cap seal in Stirling engine.

Under dry friction conditions, wear of the C-ring mating surface and excessive contact pressure of the Stirling engine piston rod cap seal can reduce its sealing performance and service life. Based on the modified Archard wear model, its thermal coupling-wear mathematical model was established and performance analysis was carried out. Based on the established simulation model, the model's predictions were first made using a BP neural network (BPNN) optimized by a particle swarm optimization (PSO). On this basis, the grey relational method was used to optimize the design of the working condition parameters with the minimum average wear rate Wr and the maximum contact pressure Pcmax as the optimization objectives, and the best optimization combination scheme was derived. Finally, the optimized design method presented in this paper was verified to exhibit superior performance using the seal test bench. The results show that the BPNN optimized by PSO the coefficient of determination R2 of Wr and Pcmax higher than 90 %, the root mean square error (RMSE) lower than 0.5, the model prediction accuracy high and the combination of regression design verifies that the BPNN prediction is real and effective. Through the grey relational method, it is found that the work pressure A1 is the most significant for the multi-objective response, and the significance of the relative operating speed B1, ambient temperature D1, and friction factor C1 decreases in the order of significance: when A1 = 7 MPa, B1 = 2.5 m/s, C1 = 0.11, and D1 = 120 °C, the Wr is 5.07e−6 (mm3/s), and the Pcmax is 22.406 MPa, which is higher than that of orthogonal experimental method and the original data, Wr decreases by 8.75 % and 18.33 %, and Pcmax increases by 15.50 % and 35.88 %, respectively, indicating that the grey relational method is superior in terms of comprehensive performance, and the sealing performance and service life of the optimized design method in this paper are verified by the experiment. [ABSTRACT FROM AUTHOR]